Composite material fiber placement path main guide line generation method, device and storage medium

Abstract

The application provides a composite material fiber laying path main guide line generation method, equipment and storage medium. The method comprises the following steps: obtaining a current state space comprising path point information and a facet information of a main guide line to be generated, and a current action space for interpolation between a candidate angle direction and a candidate geodesic line direction; determining that the current action space is retained and iteration is not ended based on a normalized minimum turning radius reward value and a maximum angle deviation reward value obtained after the current action space is executed; generating a next path point and a new state space of the next path point according to the current action space, updating the new state space as the current state space, and returning to the starting step until the corresponding extended path is determined as the main guide line at the end of iteration. In this way, by comprehensively considering the minimum turning radius and the maximum angle deviation, the possibility of the occurrence of the main guide line gap and the possibility of the occurrence of the wrinkle can be reduced, and the feasibility and stability of the path can be enhanced.

Description

Technical Field

[0001] This invention relates to the field of aircraft composite material manufacturing technology, and in particular to a method, equipment and storage medium for generating master guide lines for composite material fiber layup paths. Background Technology

[0002] In the composite material layup process, the fiber placement path planning directly determines the geometry and arrangement quality of the fiber bundles on the curved surface, and is a key factor affecting forming efficiency, defect rate, and material utilization. In industrial practice, the layup path uses the main guide line as a framework, and its direction directly determines the global orientation field and fiber angle distribution, and affects the stability and filling integrity of parallel offset.

[0003] The related technology's geodesic extension generation method uses a high-quality patch mesh for the discrete mold surface; the laying direction is determined according to the structural stress; following the laying direction, the process moves from the starting point to the surface boundary, with the curvature set to zero. Each step increases the maximum angle deviation until a geodesic is generated, which serves as the main guide line.

[0004] However, as the geodesic extension generation method increases with the maximum angle deviation, although it can reduce the possibility of gaps, the turning radius will also be larger, which will cause wrinkles in the main guide line and thus affect the material properties. Summary of the Invention

[0005] This application provides an improved method, apparatus, and storage medium for generating master guide lines for composite material fiber placement paths.

[0006] This application provides a method for generating a master guide line for composite material fiber placement paths, including: Step A: Obtain the current state space containing path point information and patch information of the main guide line to be generated, and the current action space for interpolation between candidate fixed angle directions and candidate geodesic directions; the interpolation is used to generate the extended path. Step B: When executing the current action space, at least according to the reward function of the minimum turning radius reward and the reward function of the maximum angle deviation reward, the normalized minimum turning radius reward value and the maximum angle deviation reward value are obtained. Step C: Based on the minimum turning radius reward value and the maximum angle deviation reward value, determine whether to retain the current action space; if not, return to step A to continue execution; if yes, proceed to step D. Step D: Determine whether the iteration termination condition for the target endpoint has been reached; if not, generate the next path point and a new state space for the next path point according to the current action space, update the new state space to the current state space, and return to step A to continue execution until the iteration termination condition for the target endpoint is reached, and determine the extended path corresponding to the end of the iteration as the main guide line; the main guide line is used as the reference for the composite material fiber laying path.

[0007] Furthermore, obtaining the normalized minimum turning radius reward value and maximum angle deviation reward value based at least on the reward function of the minimum turning radius reward and the reward function of the maximum angle deviation reward includes: At least the minimum turning radius reward value and the maximum angle deviation reward value are normalized and summed to obtain the normalized minimum turning radius reward value and maximum angle deviation reward value; The step of determining whether to retain the current action space based on the minimum turning radius reward value and the maximum angle deviation reward value includes: At least the minimum turning radius reward value and the maximum angle deviation reward value are linearly weighted and summed to obtain the total reward value for a single step; Based on the total reward value for a single step, determine whether to retain the current action space.

[0008] Furthermore, the path point information includes the local curvature of the path point where the main guide line to be generated is located, the tangential direction vector between the current path point and the previous path point, the laying direction, and the geodesic path direction. The patch information includes the normal vector of the current patch and the normal vector of the next patch to pass through.

[0009] Furthermore, obtaining the normalized minimum turning radius reward value and maximum angle deviation reward value based at least on the reward function of the minimum turning radius reward and the reward function of the maximum angle deviation reward includes: Based on the reward functions for minimum turning radius, maximum angle deviation, and path smoothness, normalized minimum turning radius reward values, maximum angle deviation reward values, and path smoothness reward values ​​are obtained. The step of determining whether to retain the current action space based on the minimum turning radius reward value and the maximum angle deviation reward value includes: At least the minimum turning radius reward value, maximum angle deviation reward value, and path smoothness reward value should be normalized and summed to obtain the normalized minimum turning radius reward value, maximum angle deviation reward value, and path smoothness reward value.

[0010] Furthermore, the minimum turning radius reward value is determined based on a reward mechanism where the larger the turning radius of the path segment before and after the path point, the higher the reward. The maximum angle deviation reward value is determined based on a reward mechanism where the smaller the angle deviation of the path segments before and after the path point, the higher the reward. The path smoothness reward value is determined based on a reward mechanism that increases the smoothness of the main guide line in maintaining directional continuity between adjacent path segments.

[0011] Furthermore, the target endpoint includes a boundary terminal and a further cumulative endpoint for each row; the iteration termination condition for determining whether the target endpoint has been reached includes: Does the boundary endpoint fall on the boundary of the curved mesh? And / or, The iteration termination condition for determining whether the target endpoint has been reached includes: Has the target cumulative steps been reached? And / or, The target endpoint includes a further cumulative endpoint for each row; the iteration termination condition for determining whether the target endpoint has been reached includes: the difference between the cumulative endpoint of the current cumulative step and the boundary endpoint is less than a threshold.

[0012] Furthermore, generating the next path point and its new state space according to the current action space includes: The deep reinforcement learning network is trained using the state space, action space, and reward function. After determining the current action space and executing the current action space according to the current action space, the average cumulative reward of the extended path in this round is obtained. If the average cumulative reward value of the extended path in this round is higher than the average cumulative reward value of the extended path in the previous round, the adjustment strategy for this round is retained, and the process returns to step A to continue until the iteration ends when the target endpoint is reached. At that point, the extended path corresponding to the end of the iteration is determined as the main guiding line. If the average cumulative reward value of the extended path in this round is lower than the average cumulative reward value of the extended path in the previous round, then the adjustment strategy of this round is discarded, and the next path point and the new state space of the next path point are generated according to the current action space. The new state space is updated to the current state space, and the process returns to step A to continue execution until the iteration end condition of reaching the target endpoint is reached. At that time, the extended path corresponding to the end of the iteration is determined as the main guiding line.

[0013] Furthermore, the method also includes obtaining candidate angular directions and candidate geodesic directions in the following manner: Project the previous direction onto the adjacent face and extend it symmetrically to obtain the target endpoint, adjacent face, new outgoing edge, and candidate geodesic directions; Project the global reference direction onto the current patch tangent plane and rotate it around the angle, then find the intersection within the current patch to obtain the target endpoint and the candidate fixed-angle direction.

[0014] This application provides a method for generating a master guide line for a composite material fiber placement path, the method comprising: Get the current actual state; The current actual state is input into the trained main guide line generation model, and the preferred main guide line is output; the trained main guide line generation model is generated by training the main guide line generation method for composite material fiber placement path as described above.

[0015] This application provides an electronic device including one or more processors for implementing the method described in any of the preceding claims.

[0016] This application provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the method described in any of the preceding claims.

[0017] This application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the method described in any of the preceding claims.

[0018] In some embodiments, the composite material fiber layup path main guide line generation method of this application, by comprehensively considering the minimum turning radius and the maximum angular deviation, can reduce the possibility of gaps and wrinkles appearing in the main guide line, thereby enhancing the feasibility and stability of the path. Furthermore, it can automatically generate high-quality main guide lines without tedious manual parameter adjustments, thereby improving the efficiency of main guide line design and the efficiency and forming quality of the overall path planning. Attached Figure Description

[0019] Figure 1 The diagram shown is a flowchart illustrating the method for generating the main guide line of the composite material fiber placement path provided in an embodiment of this application. Figure 2 As shown Figure 1 The diagram shown illustrates the specific process of generating the master guide line for composite material fiber placement paths, which includes a reinforcement learning network. Figure 3 As shown Figure 1 The diagram shows the framework of the reinforcement learning algorithm for the method of generating the main guide line for the composite material fiber placement path. Figure 4 The diagram shown is a schematic representation of the application process of the composite material fiber placement path master guide line generation method provided in this application embodiment; Figure 5 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0021] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0022] To address the technical problem that wrinkles occur in the main guide lines generated by geodesic extension generation methods in related technologies, affecting material properties, this application provides a method for generating main guide lines for composite material fiber layup paths. By using at least a minimum turning radius and a maximum angular deviation, the possibility of gaps appearing in the main guide lines is reduced, while the possibility of wrinkles also decreases.

[0023] Figure 1 The diagram shown is a flowchart illustrating the method for generating the main guide line of the composite material fiber placement path provided in an embodiment of this application.

[0024] like Figure 1 As shown, the method for generating the main guide line of the composite material fiber placement path may include, but is not limited to, the following steps 110 to 160: Step 110: Obtain the current state space containing path point information and patch information of the main guide line to be generated, and the current action space for interpolation between the candidate fixed angle direction and the candidate geodesic direction; interpolation is used to generate the extended path.

[0025] The aforementioned patch information provides detailed information about the local curved surface.

[0026] The aforementioned action space refers to the interpolation operation between a fixed angular direction and a geodesic direction, thereby generating a smooth and compliant extension path under different constraints. In this way, the interpolation makes the transition between path points smoother, reducing potential problems caused by abrupt path changes.

[0027] It should be noted that the aforementioned master guide lines to be generated refer to the master guide line samples in the training sample set. The path point information of the master guide line samples includes at least one of spatial coordinates and related attribute data. These master guide lines to be generated provide the foundation for subsequent interpolation operations and path optimization. This application uses deep reinforcement learning to learn the mapping relationship between the current state space and the current action space of the master guide lines to be generated, thereby achieving intelligent adjustment and optimization of path point information. In this way, it is possible to generate master guide lines that meet process requirements under complex surface conditions, while improving the efficiency and accuracy of path planning.

[0028] The process of obtaining the current state space also includes the extraction and processing of local geometric features. These local geometric features include, but are not limited to, curvature, normal vectors, and boundary conditions, which can effectively assist the path generation algorithm in making decisions in complex curved surface environments. By comprehensively considering these features, it can be ensured that the generated master guide line not only meets geometric constraints but also adapts to the needs of actual wire placement processes.

[0029] The aforementioned current state space refers to the geometric information and at least one of various process-related constraints in composite material fiber placement path planning. These constraints may include, but are not limited to, at least one of the following: material physical properties, motion limitations of the fiber placement equipment, and quality requirements of the final product.

[0030] Step 120: While executing the current action space, obtain the normalized minimum turning radius reward value and maximum angle deviation reward value based at least on the reward function of minimum turning radius reward and the reward function of maximum angle deviation reward.

[0031] The minimum turning radius bonus value is used to reduce the likelihood of gaps appearing in the main guide wire. The maximum angular deviation bonus value is used to reduce the likelihood of wrinkles appearing. Thus, the maximum angular deviation bonus value is used to constrain the variation in path direction, keeping it within a reasonable range, thereby ensuring the stability and consistency of the wire laying process.

[0032] The two reward values ​​in step 120 above work together and, after normalization, can achieve a balance in multi-objective optimization, further improving the quality and efficiency of path generation for the main guiding line.

[0033] Step 130: Based on the minimum turning radius reward value and the maximum angle deviation reward value, determine whether to retain the current action space; if not, it means that the current action space is not retained, and return to step 110 to continue execution; if yes, it means that the current action space is retained, and proceed to step 140.

[0034] Step 140: Determine whether the iteration termination condition for the target endpoint has been met. If not, it means that multiple iterations have not ended, and proceed to step 150; if yes, it means that multiple iterations have ended, and proceed to step 160.

[0035] Step 150: Based on the current action space, generate the next path point and the new state space of the next path point, update the new state space to the current state space, and return to step 110 to continue execution until the iteration end condition of reaching the target endpoint is met. The extended path corresponding to the end of the iteration is determined as the main guide line; the main guide line is used as the reference for the composite material fiber laying path.

[0036] Step 160: Determine the extension path corresponding to the end of the iteration as the main guide line; the main guide line is used as the reference for the composite material fiber laying path.

[0037] The aforementioned master guide line refers to the path line that serves as a reference during the fiber placement process of composite materials. Thus, it not only determines the overall direction of fiber placement but also directly affects the final material's performance.

[0038] In this paper, deep reinforcement learning is used to learn the mapping relationship between the current state space and the current action space of the master guide line to be generated. This process is repeated multiple times until the iteration ends at the target endpoint. At this point, the extended path corresponding to the end of the iteration is selected as the preferred master guide line. Therefore, the extended path corresponding to the end of the iteration is determined as the master guide line, and a trained master guide line generation model is obtained for subsequent independent deployment and application.

[0039] The simulation scenario for generating the main guide line of the composite material fiber layup path in this paper is based on the construction of a triangular meshed layup surface, and its input information is jointly provided by the point data table Point Table.csv and the face data table Face Table.csv.

[0040] The Point Table.csv file contains N rows of valid data, where N represents the number of valid points in the point data table. The i-th row contains records... Subscript Represents the first point in the point data table. One sampling point; Indicates the first A column vector of coordinates of sampling points in three-dimensional space, with superscript... Indicates transpose. Represents a three-dimensional real vector space; Indicates the first The number of each sampling point, subscript This indicates that the number is related to the first Row data correspondence. First, according to... Establish a mapping from point number to vertex index. ,in, A function representing the mapping from point numbers to array indices; a set This indicates that the vertex array uses a storage index range starting from 0; subscript This represents the index of the point's storage location in the vertex array. Then, sorting the vertices by index yields... in Indicates the index of the vertex array as The column vector of three-dimensional coordinates of the vertices, with subscripts It is emphasized that these are array indices, not raw row numbers; For index The corresponding coordinate components. This leads to the definition of the set of all vertices. , where the symbol To represent a subset relation, therefore This represents a 3D point set consisting of N vertex coordinates. The above mapping corresponds to a vertex index dictionary: its key is... The value is the index of the vertex in the array; at the same time, Treating it as a row index of the point data table, thus obtaining the vertex lookup dictionary. in This represents a dictionary that allows direct lookup of the 3D coordinates of a point by its index; therefore, given any point index... You can pass Access its three-dimensional coordinates.

[0041] The Face Table.csv file contains M rows of valid data, where the index M represents the number of valid triangular faces in the face data table. A row corresponds to a triangular facet, where the index... Indicates the first Each patch, this row field includes , , , , , , , , , , , , , , , , , , , , in, Indicates the first The number of each piece; Represents a piece of dough The Edge number, subscript Represents edge number and subscript This indicates that the field belongs to a patch. ; Represents a piece of dough The The vertex numbers of the two endpoints of the edge, where the overlined ones are... This indicates the first or second endpoint of the edge (this only distinguishes the endpoints and does not represent the direction); Represents a piece of dough The Is the edge marked as a boundary edge of the model? Represents and patches The The number of the face adjacent to the edge (if there is no adjacent face, it is recorded as 0); Represents a piece of dough The normal vector of is the component values ​​of the vector in the three-component directions of the three-dimensional coordinate system, where the subscripts are... Used to distinguish the three component indices; Represents a piece of dough Is it a marker for a boundary patch?

[0042] For the The six endpoints of the three sides of each triangular facet are numbered and their union is taken. , , , , , subscript This indicates the facet corresponding to the set of endpoints. ; Represents a piece of dough The set of endpoint numbers of the three edges. And using a vertex lookup dictionary. After deduplicating the endpoint numbers, we obtain the vertex set. , Among the symbols Representing subset relations; Represents a set The cardinality, therefore Represents a piece of dough It consists of exactly 3 distinct vertices; Represents a piece of dough The set of vertices with three vertex numbers. Let it be denoted as... subscript pairs Indicates a dual index: Indicates the noodle piece number. This indicates the first facet. There are vertices; therefore Represents a piece of dough The Each vertex is numbered. This is achieved through mapping. The coordinates of the three vertices of the face in geometric space can be obtained. , ,in Represents a piece of dough The A column vector of three-dimensional coordinates of vertices; subscripts Indicates the part to which the subscript belongs. This represents the vertex index within the face. This leads to the face vertex dictionary. in The table "a dictionary for quickly looking up the coordinates of the three vertices of a facet by its number" is used to quickly read the vertex coordinates of any triangular facet in subsequent algorithms.

[0043] For each piece Read the normal vector components directly from the facet table Let the original vector be... and normalize ,in, The superscript "~" indicates the unnormalized original vector, therefore Represents a piece of dough The original normal vector; Indicates will According to Euclidean norm The normalized unit normal vector; Represents a piece of dough The magnitude of the original normal vector. Based on this, define the dictionary of patch normal vectors. in This dictionary represents the unit normal vector queried by patch number, used to provide normal information for the local orientation control and paving angle constraints of the master guide line.

[0044] At the same time, the centroid of each triangular facet can be calculated using the vertex coordinates. subscript This indicates the noodle number, therefore Represents a piece of dough Geometric centroid coordinates; coefficients This represents taking the arithmetic mean of the three vertices. This leads to the construction of a dictionary of the center points of the face. ,in The dictionary that indicates the centroid coordinates of a facet by its number is used to represent the geometric location of that facet.

[0045] Regarding topological relationships, for the first Let each facet be represented by its three edges as an unordered set of endpoints. and the adjacent face numbers in the face table. Pairing them together yields triples. , in Represents a piece of dough Along its first Number the adjacent faces of the strip, and satisfy the following conditions: This indicates that the edge is a model boundary edge. When it means that the edge is numbered If the faces are adjacent, organize the above triples into an adjacency dictionary according to the face numbers. in This represents a dictionary query that retrieves the three edges and information of the corresponding adjacent faces based on the face number. The three edges of the face and their corresponding adjacent face numbers can be obtained, thus from... Whether or not to distinguish between boundary edges and internal edges.

[0046] The above vertex lookup dictionary Patch vertex dictionary Surface normal vector dictionary Dictionary of the center point of a face Adjacency Dictionary Together they support the continuous extension of the main guiding line on the triangular mesh and the local orientation update.

[0047] In this example, from the set of non-boundary patches Select a starting facet This serves as the starting facet for the main guiding line. The subscript int indicates "interior". This represents the set of all non-boundary patch numbers; Index variable representing any facet; Indicates the number is The subscript indicating whether a face is a boundary face. This indicates that the mark is related to the face. Correspondence; condition Represents a piece of dough Non-boundary surface; The subscript 0 indicates the initial face selected. (Starting face) The geometric center is defined as And it is defined as the starting point of the main leader line. Among them, The subscript 0 indicates the initial point of the master leader line; Indicates the starting facet The three vertex coordinate column vectors, where This indicates that these vertices belong to the face. , and subscript The three vertices used here to distinguish the face; coefficients This indicates that the arithmetic mean of the coordinates of the three vertices is taken, thus Represents a piece of dough The centroid coordinates are determined. The normal vector components of the surface are read from the surface normal vector dictionary, and the surface is denoted as... The normal vector is in, Indicates the starting facet normal vector, subscript This indicates the facet corresponding to the normal vector. ; Represents a piece of dough The component values ​​of the normal vector in the three component directions, where the subscripts are... Indicates taken from a sheet of dough The record, in the subscript Here, the normal vector is represented as... Component index in direction; This indicates transpose, used to organize the three components into a column vector. Combined with a preset target placement angle... Utilizing direction Functional Obtain the unit tangential direction located in the tangential plane of the patch. This determines the initial point of the main guiding line. relative to the initial direction ,in, This indicates the preset target laying angle parameters; For the direction generation function based on angle constraints, its subscript Indicates "angle", therefore Indicates the normal direction of the surface Within the defined tangent plane, according to the angle Operations to generate direction vectors; The subscript 0 indicates the initial direction of the master guide line; constraints express It is a three-dimensional vector, and in the context of this article... Indicates the location of the face. The unit tangential direction vector within the tangential plane.

[0048] In this embodiment, the current state space is defined as follows: The current state space is constructed as a real vector space of length 20. ,in, Indicates the first The state vector corresponding to each path point, index This indicates that the state changes with the path point number. This represents a 20-dimensional real vector space.

[0049] Among them, the local curvature at the location of the path point Let be the first-dimensional scalar of the current state space, where the subscript is ? Indicates the numbering sequence along the main leader line. Path points, The local curvature value representing the location of the path point (normalized to) ), corresponding to the first dimension of the current state space, The normal vector of the triangle facet where the current path point is located The set of normal vectors of its adjacent face The average included angle is normalized to obtain, where, Indicates containing the first The normal vector of the current triangle facet at each path point; set index This indicates that all faces adjacent to the current face are enumerated. Indicates the first The set of adjacent face indices of the face containing each path point, with subscripts... This indicates that the set changes as the path points are located; Indicates that the index is The normal vector of the adjacent face.

[0050] Specifically, , , ,in Indicates the current face and the first face. The normal angle between adjacent facets (unit: degrees); The hyphen "-" indicates that the average is calculated over the neighborhood. Indicates the first The average angle between the neighborhood normal vectors at each path point (unit: degrees); Represents a set The number of elements (number of adjacent faces); Indicates that the value is truncated to interval to ensure The domain is valid; the symbol This represents the dot product of vectors. Therefore... Indicates the first The local curvature scalar at the path point is normalized to Interval.

[0051] The tangential direction vectors of the current path point and the previous path point are denoted as follows: and These correspond to the 2nd to 4th dimensions and the 5th to 7th dimensions of the current state space, respectively. Indicates the first Tangential direction at each path point; subscript Indicates the previous path point (the first path point) Tangential direction at (points); Represents a three-dimensional spatial vector.

[0052] Laying direction This corresponds to the 8th to 10th dimensions of the current state space, where the superscript... Indicates "target", therefore Indicates the first The target placement direction at each path point. The calculation method is as follows: first, take the global reference vector... subscript This indicates that the reference direction is taken in the global coordinate system. Axial direction; superscript Indicates transpose. Projected onto the current facet cutting plane to obtain , ,in The superscript "~" indicates an unnormalized quantity; therefore Indicates the first At each path point, the projection vector of the global reference direction onto the current tangent plane; This indicates the reference direction of the tangent plane after normalizing the projection vector; This represents the magnitude of the vector. Then, it revolves around the unit normal vector. Rotation angle Received ,in superscript " " indicates unitization, therefore Indicates the first The unit normal vector of the current face at each path point; Indicates rotation around the axis Rotation angle rotation matrix; This represents the angular parameter indicating rotation about the normal; therefore... Indicates the first The target placement direction at each path point.

[0053] Geodesic path direction These are the 11th to 13th dimensions of the current state space, where the superscript... It represents "geodesic geodesy", therefore Indicates the first The geodesic line extension direction at each path point is determined by the direction at the previous moment. Projecting onto the current tangent plane yields in The superscript "~" indicates an unnormalized quantity, therefore Indicates the direction of the previous direction Projected to the The vector following the current tangent plane at the point; Represents the magnitude of the projection vector; A threshold to prevent division by zero.

[0054] When the preceding slice normal vector This corresponds to the 14th to 16th dimensions of the current state space, where the subscripts... Indicates the first The current patch corresponding to each path point, therefore Indicates the first The normal vector of the triangular facet where each path point is located.

[0055] Smoothness index between several adjacent path points It is a 1-dimensional scalar, corresponding to the 17th dimension of the current state space, where the subscript... Indicates the first The smoothness is obtained by statistically analyzing the path points at the current point; therefore... Indicates the first The path smoothness index is defined by the path smoothness index within the neighborhood of each path point. The path points were calculated, where This represents the number of path points included in the statistics. Adjacent path points are denoteed as... The coordinates of the path points are ,in Indicates the first The three-dimensional coordinates of each path point; subscript arrive Indicated by Backtracking from the endpoint There are several path points. For each intermediate point... Construct two adjacent unit direction vectors And determine the angle of change of direction. The superscript "-" indicates the incident direction from the previous path point to the current point. Represents path segment The unit direction; the superscript "+" indicates the outward direction from the current path point to the next path point, therefore Represents path segment The unit direction; Indicates the first The turning angle at each midpoint (unit: degrees). Take the average angle. , in The superscript "-" indicates the average value of the turning angle, therefore Indicates the first Each point represents the average turning angle of the current point; The number of average angles involved in the calculation; Used to guarantee Falling Interval.

[0056] The normal vector passing through the next face The 18th to 20th dimensions of the current state space, where the superscript... It means "next face", therefore Indicates from the first The normal vector of the next facet that a path point is about to enter.

[0057] Therefore, the overall current state space can be represented as It has 20 dimensions. This indicates transpose, used to concatenate three-dimensional vectors into a column vector by their components; This indicates that vectors are concatenated in a predetermined order. Therefore, the current state space... according to The formal organization depicts the local geometric and path evolution characteristics of the main guide line during the laying process.

[0058] In this embodiment, the current action space is defined as a one-dimensional continuous action space. That is, the intelligent agent at every moment Output one-dimensional continuous scalar action ,in, This indicates that the candidate geodesic directions are used completely. This indicates that the candidate fixed-angle direction is fully adopted. Let the normal vector of the current triangular facet be denoted as . The candidate fixed-angle direction obtained from the target laying angle is The candidate geodesic direction obtained by projecting the tangential direction from the previous moment and the current normal vector is: The actual laying direction in this step is determined by both. Perform linear interpolation and normalization to obtain in This refers to the actual laying direction of the environment within the current triangular facet used to solve for the intersection of the ray and the boundary, and then generate the next path point.

[0059] In this embodiment, by minimizing the turning radius and maximizing the angular deviation, the likelihood of gaps and wrinkles appearing in the main guide line can be reduced, thereby enhancing the feasibility and stability of the path. Simultaneously, high-quality main guide lines can be automatically generated without tedious manual parameter adjustments, thus improving the efficiency of main guide line design, overall path planning efficiency, and forming quality. This is particularly suitable for the automatic wire placement of aerospace composite material components with large curvature, rapid curvature changes, and complex multi-boundary surfaces.

[0060] like Figure 1 As shown, steps 120 and 130 above can be implemented using at least one of the following embodiments to determine whether to retain the current action space: In a first optional embodiment, step 121 involves normalizing and summing the minimum turning radius reward value and the maximum angle deviation reward value to obtain the normalized minimum turning radius reward value and maximum angle deviation reward value.

[0061] Accordingly, in step 1311, at least the minimum turning radius reward value and the maximum angle deviation reward value are linearly weighted and summed to obtain the total reward value for a single step.

[0062] Step 1312: Based on the total reward value for a single step, determine whether to retain the current action space.

[0063] The above allows setting different weights for the minimum turning radius reward value and the maximum angle deviation reward value, as well as a threshold for the total reward value per step. If the total reward value per step exceeds the threshold, the current action space is retained; if the total reward value per step does not exceed the threshold, the parameters need to be readjusted or another action space needs to be selected to ensure the quality and efficiency of path generation. This not only improves the accuracy of the main guideline path planning but also allows for the selection of a better main guideline path.

[0064] In this embodiment, by linearly weighting and summing at least the minimum turning radius reward value and the maximum angle deviation reward value, the impact of the minimum turning radius and the maximum angle deviation on path generation can be effectively balanced. This not only reduces the likelihood of gaps in the main guide line but also reduces the likelihood of wrinkles. Simultaneously, the total reward value per step can determine whether the current motion space meets the requirements for path generation of the main guide line. By setting a reasonable threshold, the path generation process can be further optimized.

[0065] As an optional embodiment of this application, the path point information includes the local curvature of the path point where the main guide line to be generated is located, the tangential direction vector between the current path point and the previous path point, the laying direction, and the geodesic path direction.

[0066] The combination of the laying direction and the geodesic path direction helps ensure that the composite material maintains ideal stress distribution characteristics during the laying process. Geometric constraints can be generated for the master guide line based on the current patch normal vector and the normal vector of the next patch to pass through.

[0067] The patch information includes the normal vector of the current patch and the normal vector of the next patch to pass through.

[0068] In this embodiment, by combining path point information and patch information for analysis, the direction of the main guiding line can be predicted more accurately, and potential geometric conflicts can be effectively avoided. Simultaneously, by utilizing local curvature and tangential direction vectors, the smoothness between path points can be further optimized, thereby improving the continuity and stability of the overall path.

[0069] Combination Figure 1 As shown, in the second embodiment, step 1221 involves obtaining normalized minimum turning radius reward values, maximum angle deviation reward values, and path smoothness reward values ​​based on the reward functions for minimum turning radius reward, maximum angle deviation reward, and path smoothness reward.

[0070] Accordingly, in step 1321, based on the minimum turning radius reward value, the maximum angle deviation reward value, and the path smoothness reward value, it is determined whether to retain the current action space.

[0071] As an optional embodiment, the minimum turning radius reward value is determined based on a reward mechanism where a larger turning radius on the path segments before and after the waypoint results in a higher reward. Thus, the turning radius and reward are positively correlated. A larger turning radius helps reduce the mechanical stress on the equipment during path switching, thereby improving operational safety and stability.

[0072] The maximum angular deviation reward value is determined based on a reward mechanism where the smaller the angular deviation of the path segments before and after the path point, the higher the reward. Thus, angular deviation and reward are negatively correlated. In this way, a smaller angular deviation ensures the consistency of the path direction, avoiding error accumulation or decreased wire-laying quality caused by abrupt changes in direction.

[0073] From the starting point to the boundary of the curved surface, where the curvature is zero, a larger maximum angular deviation reduces the likelihood of gaps appearing. A larger turning radius reduces the likelihood of wrinkles, thus minimizing wrinkles and preventing them from affecting material properties.

[0074] The path smoothness bonus value is determined based on a reward mechanism where the greater the smoothness of the main guide line maintaining directional continuity between adjacent path segments, the higher the reward. This ensures that the main guide line maintains directional continuity and exhibits a zigzag oscillation between adjacent path segments. Optimizing path smoothness not only improves the visual appearance of the main guide line but also reduces potential oscillations or jitters during subsequent yarn laying, thereby improving overall yarn laying efficiency and accuracy.

[0075] In this embodiment of the application, normalized minimum turning radius reward value, maximum angle deviation reward value and path smoothness reward value are obtained through the respective reward mechanisms of minimum turning radius, maximum angle deviation and path smoothness, which can effectively guide the optimization of turning radius, angle deviation and smoothness in the path planning process.

[0076] Furthermore, step 1321 above may include, but is not limited to: normalizing and summing the minimum turning radius reward value, the maximum angle deviation reward value, and the path smoothness reward value to obtain normalized minimum turning radius reward values, maximum angle deviation reward values, and path smoothness reward values. Thus, according to the turning radius reward function, angle deviation reward function, and path smoothness reward function, the reward function includes satisfying the minimum turning radius reward, satisfying the maximum angle deviation reward, and the path smoothness reward, and each sub-reward function is normalized and summed.

[0077] To avoid imbalance of the main term due to the different dimensions and numerical ranges of physical quantities, the sub-terms are first normalized. The normalization function is defined as follows: ,in, It represents any physical quantity or intermediate variable that needs to be normalized or have its numerical range limited; Indicates will The lower bound is 0; This indicates that the upper bound of the result is restricted to 1; This represents a normalization function designed to eliminate dimensional differences in physical quantities.

[0078] In this embodiment, the discrete time step is set to be... The current path point sequence is The target placement angle is constant. The current unit normal vector of the triangular facet is The specific design of the reward function is as follows: 1. First, define the normalization function as follows: ,in, It represents any physical quantity or intermediate variable that needs to be normalized or have its numerical range limited; Indicates will The lower bound is 0; Limit the upper bound of the result to 1; This represents a normalization function designed to eliminate dimensional differences in physical quantities.

[0079] 2. Normalized function for turning radius: The main guiding line is at the The three nearest path points at the step are ,in, For the previous path point, This is the current path point. For the next path point, calculate the turning radius using the formula for the circumcircle of three points. Three path points are set. Calculate the lengths of the three sides and the area of ​​the cross product of vectors Construct the radius of the circumcircle of these three points. Its minimum physical allowable value is Similarly, candidate angle points are constructed using the same phase length along the angle direction and the geodesic direction, respectively. Candidate geodesic points The turning radii calculated from the candidate fixed angle points are obtained respectively. and the turning radius calculated from the candidate geodesic points , denoted as the upper limit of typical clipping. The normalized form of the turning radius is defined as: .

[0080] 3. Angular deviation normalization function: The main guiding line is at the Step, based on the current triangle normal vector Angle of placement relative to the target Calculate the candidate fixed-angle direction And based on the tangential direction of the previous moment and Calculate the direction of the candidate geodesics The actions given by the intelligent agent The actual laying direction is obtained by interpolating and normalizing the two candidate directions. Record the actual laying angle deviation as Let the maximum deviation under the two extreme candidate directions be denoted as The normalized form of the angular deviation is defined as follows: .

[0081] 4. Path smoothness normalization function: The main guiding line is at the Step 1: Take a set of path point sequences for evaluation. Let vector , The included angle between two adjacent segments is defined as follows: in Indicates the path at point The angle of change of direction at that location; This indicates that floating-point errors should be prevented. Input exceeded Take the average of the included angles over the evaluation interval: ,in, This represents the average included angle that reflects the average directional change of the path in that local region. To determine the number of path points included in the calculation, the average included angle is linearly mapped to... The smoothness normalized form is obtained as follows: .

[0082] ①The turning radius reward function is: .

[0083] ②The angle deviation reward function is: .

[0084] ③ The path smoothness reward function is: .

[0085] in, This is a path validity indicator function. It takes the value 1 if the current step is successful and does not exceed the boundary, and 0 otherwise.

[0086] Taking into account the requirements of minimum turning radius, angle deviation, and path smoothness, the total reward for a single step is defined as follows: ,in, As weighting coefficients, to maintain the same order of magnitude, it is preferable to set them as follows: The training objective is to maximize the expected cumulative reward. ,in, This represents the reinforcement learning objective function with respect to policy parameters. The expected cumulative discount reward for all tracks; Representation strategy; This represents the probability distribution of the agent choosing the current action space based on the current state space; This is the discount factor. The weighting coefficient in this embodiment is taken as... In this system, the turning radius and angle deviation bonuses are treated as immediate rewards and are then added at the end of the round based on... Global smoothness reward.

[0087] In this embodiment, the path smoothness bonus value enables the main guide line to be smooth, facilitating subsequent wire laying. Simultaneously, normalization ensures that the minimum turning radius bonus value, maximum angle deviation bonus value, and path smoothness bonus value all use the same dimension, eliminating the impact of dimensional differences between these values ​​on the final result.

[0088] As an optional embodiment, the above method may also include, but is not limited to, obtaining candidate angular directions and candidate geodesic directions by employing the following steps 210 to 220: Step 210: Project the previous direction onto the adjacent face and extend it symmetrically to obtain the target endpoint, adjacent face, new outgoing edge, and candidate geodesic directions.

[0089] Step 220: Project the global reference direction onto the current patch cutting plane and rotate it around the angle. Find the intersection within the current patch to obtain the target endpoint and the candidate fixed angle direction.

[0090] Define environment parameter set This includes the current patch number. Current starting point coordinates The unit direction vector already used in the previous paragraph Global reference direction Fixed angle parameters Geodesic angle parameters The unit normal of the current facet .

[0091] Define the current action space For: a continuous scalar , representing the interpolation weight between the candidate geodesic direction and the candidate fixed-angle direction: ,in, Indicates the direction of extension of the agent's decision-making; This indicates the direction of the candidate geodesic line calculated below; This indicates the candidate fixed-angle direction calculated below; This represents the normalization formula.

[0092] Current state space Given the environment With network decision-making Then, the environment first constructs two candidate directions: The candidate geodesic direction is obtained by referencing the previous outgoing edge, projecting the previous direction onto the adjacent face, and extending it symmetrically to obtain the endpoint. Adjacent face New outgoing edges and candidate geodesic directions ; The candidate angular direction is determined by using the global reference direction. Project onto the current facet tangent plane and around Rotation angle Find the intersection within the given area to obtain the endpoint. The fixed angle direction of the candidate .

[0093] Then, using the current action space The weighted summation and normalization between the two candidate directions yields the following result. And perform a strict ray-boundary intersection along this direction within the current patch: For each edge ,remember ,calculate , ,satisfy > and When the edge is hit, the minimum value is taken. The intersection point is the end point of this step. ,in, Indicates starting from the current point Along direction The ray equation; Represents the two endpoints of an edge; This represents the edge vector; Indicates the reference normal direction; This indicates the projection ratio within the edge.

[0094] After the intersection is completed, the environment returns to the new state space. ,in, The boundary is determined by the sharing relationship of the hit edges in the face data table; if no shared face exists, it is considered a boundary termination.

[0095] In this embodiment, by symmetrically extending and projecting the candidate geodesic direction and the candidate fixed angle direction, the path discontinuity problem caused by directional deviation can be avoided, thereby improving the smoothness and stability of the overall path.

[0096] Combination Figure 1 As shown, the target endpoint includes the boundary terminal and the further cumulative endpoint for each row. Accordingly, step 140 can employ at least one of the following optional embodiments to determine whether the iteration termination condition for the target endpoint has been reached: In a first optional embodiment, the boundary endpoint is determined to fall within the boundary of the curved mesh. This boundary endpoint determination effectively prevents paths from exceeding the boundary range of the composite material.

[0097] In a second alternative embodiment, the target cumulative step count is determined. Thus, by setting a target cumulative step count, a constraint can be provided on the path length of the main guide line, ensuring that computational efficiency matches actual needs.

[0098] In a third optional embodiment, if the difference between the cumulative endpoint and the boundary endpoint of the current cumulative step is less than a threshold, it indicates that the cumulative endpoint is gradually converging to the boundary endpoint, suggesting that the path planning is gradually approaching the target. Thus, the threshold difference between the current cumulative step and the boundary endpoint allows for stable operation and output of the main guiding line result even in the face of minor errors or noise interference.

[0099] The first, second, and third optional embodiments described above can be executed independently or in combination. When multiple embodiments are combined, the accuracy of the main guide path generation can be improved.

[0100] Figure 2 As shown Figure 1 The diagram shown illustrates the specific process of generating the master guide line for composite material fiber placement paths, which includes a reinforcement learning network.

[0101] Combination Figure 1 and Figure 2 As shown, step 150 above, which generates the next path point and its new state space according to the current action space, may include, but is not limited to, steps 151 to 153 as follows: Step 151: Train the deep reinforcement learning network using the current state space, current action space, and reward function. After executing the current action space, obtain the average cumulative reward of the extended path in this round. Training the deep reinforcement learning network using the current state space, current action space, and reward function allows the deep reinforcement learning network to learn the mapping relationship between the state space and the action space.

[0102] In some examples, deep reinforcement learning networks are constructed using SAC (Soft Actor-Critic).

[0103] In other examples, deep reinforcement learning networks are implemented using TD3 (Twin Delayed Deep Deterministic Policy Gradient). In other examples, deep reinforcement learning networks are constructed using the A2C (Advantage Actor-Critic algorithm).

[0104] In other examples, deep reinforcement learning networks are constructed using ACER (Actor-Critic with Experience Replay).

[0105] In other examples, deep reinforcement learning networks are constructed using DDPG (Deep Deterministic Policy Gradient).

[0106] In other embodiments, the deep reinforcement learning network is constructed using the PPO algorithm.

[0107] Figure 3 As shown Figure 1 The diagram shows the framework of the reinforcement learning algorithm for generating the master guide line of the composite material fiber placement path.

[0108] The improved PPO algorithm, based on reinforcement learning, serves as a reinforcement learning-based framework for master guide line generation in composite material fiber placement path planning. Its framework is as follows: Figure 3 As shown, the improved PPO (Proximal Policy Optimization) algorithm framework includes a policy network (Actor) and a value network (Critic). The PPO optimization objective is to cut the agent objective function. , , The current policy given to the Actor network in the current state space Next to the current action space The probability density, This is the old strategy frozen after the previous iteration; For the dominant function, From time Cumulative discount rewards The state value output by the Critic network. is the shear coefficient.

[0109] During the training phase, the Actor network adjusts its state based on the current state space. Output Action The environment generates the next path point according to the aforementioned geometric process and returns the reward. With the new state space Critic network estimates state value Both participate in the aforementioned shearing target. Iterative optimization; after training, a fixed-policy network. It interacts with the environment to make autonomous decisions, enabling the laying path to extend smoothly until the termination condition is met, while satisfying the surface geometric constraints.

[0110] Furthermore, the Actor network input state vector First, intermediate features are obtained through a three-layer feature extraction module: the first and second layers will... The layers are sequentially mapped to a 256-dimensional hidden space. After each layer, layer normalization, ReLU activation, and Dropout are applied to enhance nonlinear expressiveness and suppress overfitting. The third layer compresses the features to 128 dimensions and activates them with ReLU to obtain a shared feature vector. ReLU represents the activation function that enhances nonlinear learning capabilities; Dropout randomly discards some neurons to make training more stable. Based on this, a linear output head is used to... Mapped to action mean vector After The function restricts it to Interval: .in, Indicates the weights of the linear layer; This represents the bias of the linear layer. Action logarithm standard deviation vector. The standard deviation is obtained by independently modeling it as a learnable parameter and performing exponential and truncation operations. Thus, a Gaussian policy is constructed at each time step. The value network Critic uses the same state vector. The input is passed through four fully connected hidden layers, with the number of hidden units being... , , , Each layer is followed by ReLU activation, and finally mapped to a one-dimensional state value through a linear output layer. By combining the aforementioned Actor-Critic structure with the PPO shearing optimization objective, the learning of the layup path direction selection strategy is achieved. Specifically, the number of hidden units is 256, 256, 256, and 128, respectively.

[0111] Based on this, an improved PPO algorithm is used to construct a deep reinforcement learning network. The network is trained using the state space, action space, and reward function, enabling it to learn to adjust path generation strategies based on the state and extend the path until the termination condition is met.

[0112] In each training process, firstly, an arbitrary state space is randomly initialized. In this state space, from Sampling control actions in the network And obtain the reward at this time. Then the state transitions to the next state space. Next, , , and Each sample is stored as a sample in the experience replay pool. Finally, a small batch of samples is drawn from the experience replay pool to train the network, and the network is updated using stochastic gradient descent. In this way, after each training cycle, the updated samples can be used... The network outputs the optimal master bootstrap path.

[0113] The output target and data structure are defined as follows: Suppose that the policy network trained in step S4 generates N consecutive paths, each path denoted as: , ,in, These are the coordinates of the start and end points of the path segment, respectively. Number the triangular facet containing this path segment ( The complete set of master guide path points is then defined as: And a set of path segments: The starting point of the number. Determined by the selected starting point, the end point This is the last valid intersection point of the path at the surface boundary or target surface.

[0114] The criteria for determining the rationality of a path are defined as follows: To ensure the geometric continuity and technological feasibility of the path, the output path set is... The following constraints must be met: ① Angle deviation constraint: Adjacent path segments are continuous in direction and have no backtracking, satisfying the requirement. < ,in, This is the maximum permissible angular deviation threshold.

[0115] ② Turning radius constraint: For any three consecutive points Calculate the local turning radius ,Require ,in, This is the minimum allowed turning radius threshold.

[0116] ③ Boundary and Termination Conditions: If the endpoint Set of boundaries of curved mesh or its relationship with the target endpoint The Euclidean distance satisfies If this is achieved, the path extension task is considered complete, and the generation of the main guide line is finished.

[0117] The operability and post-output are defined as follows: ① Spatial continuity: The path point sequence is connected in the order of generation to form a continuous spatial polyline, which is used to drive the filament laying trajectory planning; ② Formatted output: The path point coordinate sequence can be exported as a matrix: , Each row corresponds to the three-dimensional coordinates of the path point, which are used for the output results.

[0118] Step 152: If the average cumulative reward value of the extended path in this round is higher than the average cumulative reward value of the extended path in the previous round, then retain the adjustment strategy for this round and continue to return to step 110 to continue execution until the iteration end condition of reaching the target endpoint is met, and the extended path corresponding to the end of the iteration is determined as the main guiding line.

[0119] Step 153: If the average cumulative reward value of the extended path in this round is lower than the average cumulative reward value of the extended path in the previous round, discard the adjustment strategy of this round, generate the next path point and the new state space of the next path point according to the current action space, update the new state space to the current state space, and return to step 110 to continue execution until the iteration end condition of reaching the target endpoint is reached, and determine the extended path corresponding to the end of the iteration as the main guiding line.

[0120] In this embodiment, continuous iterative optimization effectively improves the accuracy and efficiency of path planning. During each iteration, the system dynamically adjusts policy parameters based on the current state and action spaces to ensure that the generated extended paths better meet actual needs. Simultaneously, the reward function further guides the reinforcement learning network towards the target direction, avoiding getting trapped in local optima. Finally, after reinforcement learning or training, upon reaching the preset iteration termination condition, the determined main guiding line is output as the optimal path, serving as the benchmark for composite material fiber placement paths and providing a reliable foundation for the flat-lay path in composite material fiber placement processes.

[0121] Based on the same inventive concept as the methods described above, this application also provides a method for generating the main guide line of a composite material fiber placement path. Figure 4 The diagram shown is an application flow diagram of the composite material fiber placement path main guide line generation method provided in the embodiments of this application. The above method may include, but is not limited to, the following steps 310 to 320: Step 310: Obtain the current actual state.

[0122] Step 320: Input the current actual state into the trained main guide line generation model and output the preferred main guide line; the trained main guide line generation model is generated by training the above composite material fiber laying path main guide line generation method.

[0123] The aforementioned optimized master guide lines can significantly improve the accuracy and efficiency of composite material fiber placement. By combining the current actual state with the experience accumulated during model training, human intervention can be effectively reduced, while ensuring that the generated master guide lines are more accurate and effective, facilitating subsequent fiber placement.

[0124] In this embodiment, a trained master guide line generation model can be used to efficiently handle complex composite material fiber placement path planning problems. This trained master guide line generation model can quickly obtain the optimal master guide line based on the actual input state, thereby significantly improving the accuracy and efficiency of path generation.

[0125] Based on the same inventive concept as the above method, this application also provides a composite material fiber placement path main guide wire generation device, which may include the following modules: The information acquisition module is used to acquire the current state space containing path point information and patch information of the main guide line to be generated, as well as the current action space for interpolation between the candidate fixed angle direction and the candidate geodesic direction; the interpolation is used to generate the extended path. The reward value determination module is used to obtain the normalized minimum turning radius reward value and maximum angle deviation reward value, at least based on the reward function of the minimum turning radius reward and the reward function of the maximum angle deviation reward, when executing the current action space; the reward function is the minimum turning radius reward and the maximum angle deviation reward that have been at least normalized. The current action space decision module is used to determine whether to retain the current action space based on the minimum turning radius reward value and the maximum angle deviation reward value; if not, it returns to the information acquisition module to continue execution; if yes, it executes the main guide line determination module. The main guide line determination module is used to determine whether the iteration end condition of the target endpoint has been reached; if not, it generates the next path point and a new state space of the next path point according to the current action space, updates the new state space to the current state space, and returns to the information acquisition module to continue execution until the iteration end condition of the target endpoint is reached, and determines the extended path corresponding to the end of the iteration as the main guide line; the main guide line is used as the reference for the composite material fiber laying path.

[0126] As one embodiment, the composite material fiber placement path master guide line generation device further includes: obtaining candidate fixed-angle directions and candidate geodesic directions in the following manner: Project the previous direction onto the adjacent face and extend it symmetrically to obtain the target endpoint, adjacent face, new outgoing edge, and candidate geodesic directions; Project the global reference direction onto the current patch tangent plane and rotate it around the angle, then find the intersection within the current patch to obtain the target endpoint and the candidate fixed-angle direction.

[0127] Each module of the above-mentioned device corresponds to the steps of the above-mentioned method. The specific implementation process of the function and role of each module in the above-mentioned device can be found in the implementation process of the corresponding steps in the above-mentioned method, which can achieve the same technical effect, and will not be repeated here.

[0128] This application provides an electronic device, including one or more processors, for implementing the composite material fiber layup path master guide line generation method as described above.

[0129] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, or a PDA (Personal Digital Assistant). The PDA can include industrial PDAs and consumer PDAs. Any electronic device that can implement the embodiments of this invention is within the scope of protection of this invention and is not limited thereto.

[0130] Figure 5 The diagram shown is a structural schematic of the electronic device 50 provided in an embodiment of this application.

[0131] like Figure 5 As shown, the electronic device 50 includes one or more processors 51 for implementing the composite material fiber layup path master guide line generation method as described above.

[0132] In some embodiments, electronic device 50 may include storage medium 59. For example, a computer-readable storage medium may store a program that can be invoked by processor 51, and may include non-volatile storage medium. In some embodiments, electronic device 50 may include memory 58 and interface 57. In some embodiments, electronic device 50 may also include other hardware depending on the specific application.

[0133] The computer-readable storage medium of this application embodiment stores a program that, when executed by processor 51, is used to implement the composite material fiber layup path master guide line generation method described above.

[0134] This application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the method described in any of the preceding claims.

[0135] This application also provides a computer program stored in a computer-readable storage medium, for example... Figure 5The storage medium 59, and when the processor executes the computer program, causes the processor 51 to perform the method described above.

[0136] This application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented using any method or technology. Information may be computer-readable instructions, data structures, program modules, or other data. Examples of computer-readable storage media include, but are not limited to: phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0137] The above description is merely a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and incentive mechanism of this specification shall be included within the scope of protection of this specification.

[0138] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element qualified by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for generating a master guide line for a composite material fiber placement path, characterized in that, include: Step A: Obtain the current state space containing path point information and patch information of the main guide line to be generated, and the current action space for interpolation between candidate fixed angle directions and candidate geodesic directions; the interpolation is used to generate the extended path. Step B: When executing the current action space, at least according to the reward function of the minimum turning radius reward and the reward function of the maximum angle deviation reward, obtain the normalized minimum turning radius reward value and the maximum angle deviation reward value. Step C: Based on the minimum turning radius reward value and the maximum angle deviation reward value, determine whether to retain the current action space; if not, return to step A to continue execution; If so, proceed to step D; Step D: Determine whether the iteration termination condition for the target endpoint has been reached; if not, generate the next path point and a new state space for the next path point according to the current action space, update the new state space to the current state space, and return to step A to continue execution until the iteration termination condition for the target endpoint is reached, and determine the extended path corresponding to the end of the iteration as the main guide line; the main guide line is used as the reference for the composite material fiber laying path.

2. The method for generating the main guide line of the composite material fiber placement path as described in claim 1, characterized in that, The process of obtaining the normalized minimum turning radius reward value and maximum angle deviation reward value based at least on the reward function of the minimum turning radius reward and the reward function of the maximum angle deviation reward includes: At least the minimum turning radius reward value and the maximum angle deviation reward value are normalized and summed to obtain the normalized minimum turning radius reward value and maximum angle deviation reward value; The step of determining whether to retain the current action space based on the minimum turning radius reward value and the maximum angle deviation reward value includes: At least the minimum turning radius reward value and the maximum angle deviation reward value are linearly weighted and summed to obtain the total reward value for a single step; Based on the total reward value for a single step, determine whether to retain the current action space.

3. The method for generating the main guide line of the composite material fiber placement path as described in claim 1, characterized in that, The path point information includes the local curvature of the path point where the main guide line to be generated is located, the tangential direction vector between the current path point and the previous path point, the laying direction, and the geodesic path direction. The patch information includes the normal vector of the current patch and the normal vector of the next patch to pass through.

4. The method for generating the main guide line of the composite material fiber placement path as described in claim 1, characterized in that, The process of obtaining the normalized minimum turning radius reward value and maximum angle deviation reward value based at least on the reward function of the minimum turning radius reward and the reward function of the maximum angle deviation reward includes: Based on the reward functions for minimum turning radius, maximum angle deviation, and path smoothness, normalized minimum turning radius reward values, maximum angle deviation reward values, and path smoothness reward values ​​are obtained. The step of determining whether to retain the current action space based on the minimum turning radius reward value and the maximum angle deviation reward value includes: At least the minimum turning radius reward value, maximum angle deviation reward value, and path smoothness reward value should be normalized and summed to obtain the normalized minimum turning radius reward value, maximum angle deviation reward value, and path smoothness reward value.

5. The method for generating the main guide line of the composite material fiber placement path as described in claim 4, characterized in that, The minimum turning radius reward value is determined based on a reward mechanism where the larger the turning radius of the path segment before and after the path point, the higher the reward. The maximum angle deviation reward value is determined based on a reward mechanism where the smaller the angle deviation of the path segments before and after the path point, the higher the reward. The path smoothness reward value is determined based on a reward mechanism that increases the smoothness of the main guide line in maintaining directional continuity between adjacent path segments.

6. The method for generating the main guide line of the composite material fiber placement path as described in claim 1, characterized in that, The target endpoint includes the boundary terminal and the cumulative endpoint for each row; the iteration termination condition for determining whether the target endpoint has been reached includes: Does the boundary endpoint fall on the boundary of the curved mesh? And / or, The iteration termination condition for determining whether the target endpoint has been reached includes: Has the target cumulative steps been reached? And / or, The target endpoint includes a further cumulative endpoint for each row; the iteration termination condition for determining whether the target endpoint has been reached includes: the difference between the cumulative endpoint of the current cumulative step and the boundary endpoint is less than a threshold.

7. The method for generating the main guide line of the composite material fiber placement path as described in claim 1 or 2, characterized in that, The step of generating the next path point and its new state space according to the current action space includes: The deep reinforcement learning network is trained using the state space, action space, and reward function. After determining the current action space and executing the current action space according to the current action space, the average cumulative reward of the extended path in this round is obtained. If the average cumulative reward value of the extended path in this round is higher than the average cumulative reward value of the extended path in the previous round, the adjustment strategy for this round is retained, and the process returns to step A to continue until the iteration end condition of reaching the target endpoint is met. At that time, the extended path corresponding to the end of the iteration is determined as the main guiding line. If the average cumulative reward value of the extended path in this round is lower than the average cumulative reward value of the extended path in the previous round, then the adjustment strategy of this round is discarded, and the next path point and the new state space of the next path point are generated according to the current action space. The new state space is updated to the current state space, and the process returns to step A to continue execution until the iteration end condition of reaching the target endpoint is reached. At that time, the extended path corresponding to the end of the iteration is determined as the main guiding line.

8. The method for generating the main guide line of the composite material fiber placement path as described in claim 1 or 2, characterized in that, The method further includes obtaining the candidate angular direction and the candidate geodesic direction in the following manner: Project the previous direction onto the adjacent face and extend it symmetrically to obtain the target endpoint, adjacent face, new outgoing edge, and candidate geodesic directions; Project the global reference direction onto the current patch's tangent plane and rotate it around the angle. Find the intersection within the current patch to obtain the target endpoint and the candidate fixed-angle direction.

9. A method for generating a master guide line for a composite material fiber placement path, characterized in that, The method includes: Get the current actual state; The current actual state is input into the trained main guide line generation model, and the preferred main guide line is output; the trained main guide line generation model is generated by training the composite material fiber layup path main guide line generation method as described in any one of claims 1 to 8.

10. An electronic device, characterized in that, It includes one or more processors for implementing the composite material fiber layup path master guide line generation method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the method for generating the master guide line of the composite material fiber layup path as described in any one of claims 1 to 8.

Images